- Setting up detectron2
- Prepare a COCO dataset format json file (Ex: total_text.json)
- The dataset should have annotations such as
- width
- height
- segmentation
- image_id
- Download the dataset and create directory to place the training images (Ex: Images/Train)
- Run
python train.py
to train the model
- For inference
python inference.py --input <IMAGEDIR> --output <OUTPUTDIR> --weights-file <WEIGHTFILE> --config-file <CONFIG_FILE>